Apache Curator always uses cluster-internal addresses when reconnecting

[nightkr]

Affected Stackable version

24.3.0

Affected Apache ZooKeeper version

3.8.4

Current and expected behavior

See https://stackable-workspace.slack.com/archives/C042UDVBRQF/p1738075756767349, originally reported by @djtfmartin

Once it has connected once, Curator will pull the list of ensemble members from the ZooKeeper cluster, and use that instead of the configured connection string for reconnection attempts. This always contains the internal addresses, so clients that are not inside of the Kubernetes cluster will then fail to reconnect.

Possible solution

Ideally, fix the reported configuration so that it contains the correct addresses.

As a workaround, @gbif patched Curator to disable this functionality.

Additional context

No response

Environment

No response

Would you like to work on fixing this bug?

None